Indexing melodic sequences via Wavelet transform

Wavelet decomposition has been widely used in video and audio signal analysis. Here we propose a wavelet based method for indexing symbolic music material which is based on the Haar transform. Melodies are represented by pitch-class sequences at different levels of resolution. We show that it is possible to efficiently index long melodic sequences by means of their Haar transform at different levels without losing musical information. We present an experimental evaluation of this approach based on a subset of the RISM collection consisting in about 500 monophonic incipits. Results have been evaluated against a ground truth for the same collection for queries with diverse characteristics.